Annotating Spatial Interpretations of German Prepositions

Annotating Spatial Interpretations of German Prepositions

Annotating Spatial Interpretations of German
Prepositions
A. Müller, C. Roch, T. Stadtfeld, T. Kiss
Sprachwissenschaftliches Institut
Ruhr-Universität Bochum
Bochum, Germany
{mueller, roch, stadtfeld, tibor}@linguistics.rub.de
Traditional approaches to prepositions classify preposition
lexemes into different semantic classes like spatial, temporal or
modal ones. In this paper we reconsider the so-called spatial
prepositions and their subclassification. We propose a sense
based classification of prepositions using traditional criteria to
determine different classes of spatial-preposition interpretations.
Our work on spatial preposition senses is part of a bigger project
of developing an annotation scheme for a subset of German
prepositions.
Prepositions; spatial; preposition senses; annotation scheme;
German; classification
I.
INTRODUCTION
An investigation of the semantics of prepositions in actual
usage reveals their high potential to express relations in various
dimensions. Traditionally, the subcategorization of prepositions
is based on apparently predominant senses of the respective
preposition, so that prepositions are called spatial, temporal,
modal, and causal (among others). Recently, [1] has offered an
alternative in starting from abstract notions that are organized
in categories such as localization, quantity, manner,
accompaniment, etc. All in all comprehensive approaches that
try to systematically cover all possible senses of a preposition
are seldom found, possible exceptions being the Preposition
Project for English [2], and PrepNet for French [1].
The goal of the present study is the development of a
comprehensive as well as differentiated annotation scheme for
preposition senses, starting with an analysis of 22 simple
German prepositions.1
Currently, it serves two purposes: First, the annotation
scheme is employed for manual annotation of preposition
senses in a large corpus which has been compiled of newspaper
issues of the Swiss German “Neue Zürcher Zeitung” (NZZ)
from 1993 to 1999 (comprising 230 million words). Secondly,
the annotation scheme as well as the annotated corpus should
be used as a gold standard for the automatic classification of
preposition senses. 2
1
A scheme for the annotation of preposition sense for German prepositions
did not exist prior to our investigation. The 22 prepositions under
investigation are those simple prepositions of German that select NP as well
as bare N complements and govern a case.
2
In addition to the annotation of preposition senses, the corpus is annotated
with morphological, lexical, syntactic, and semantic markup. For further
details as well as an example of the use of the scheme cf. [3].
In the present paper, we focus on the classification of
spatial interpretations. Many studies examining the meaning of
prepositions are situated in the domains of space and time.
Spatial interpretations have received a lot of interest, as they
often tend to be regarded as the primary and most prototypical
senses. However, whether these relations should be modeled
via geometrical relations or by means of symbolic logic is still
a fundamentally controversial issue. Semanticists disagree as to
whether the local placement should be analyzed in terms of a
localization of objects or a localization of situations in their
semantic representations (cf. [4]). In addition, we find huge
differences with regard to the incorporation of a conceptual
level (so-called two-level approaches, as e.g. [5], [6]), and also
contextual or world knowledge.
We will take a stand on some of the concerns above and
also present some new reflections with respect to theoretical
questions that have not been touched yet to the best of our
knowledge. Since this paper will focus on the annotation of
spatial interpretations and – consequently – on the part of the
scheme designed to capture this dimension, we refer the reader
to [7] for a general picture.
As a starting point, we would like to state explicitly that we
talk about spatial interpretations and not about spatial
prepositions, as we adopt a perspective that we term a
relational analysis of prepositions. A major feature of this view
is that it is agnostic with respect to primary or prototypical
meanings of prepositions. Instead, we assume that the
prepositional system of a language is made up by associating
sometimes arbitrary subsets of relational meanings to
preposition lexemes. Let us illustrate this view with the
preposition über, which is often classified as a spatial
preposition and hence prototypically translated into English as
over or above. The following examples show two different
spatial interpretations in (1) and (2), and in addition, a sense
referring to a HIERARCHICAL structure (3), and an ABOUTNESS
interpretation in (4). While the first sentence is an example for
the projective vertical use of über, the second one exemplifies a
topological transit.
(1)
Die Aktion dient der Navigationsschulung über
The operation serves the navigational training above
unbekanntem Gebiet.
unknown territory
‘The navigational training above unknown territory is
facilitated by this operation.’
(2)
(3)
Die Erdgasleitung von Sibirien nach
The natural gas pipeline from Siberia to
Westeuropa verläuft über polnisches Gebiet.
Western Europe runs through Polish territory
spatial
‘The natural gas pipeline from Siberia to Western
Europe runs through Polish territory.’
purpose
Die Truppen bedrohten die Oberhoheit
The troops endangered the suzerainty
Aserbeidschans über das Gebiet.
of Azerbaijan over the territory
‘The troops endangered Azerbaijan’s suzerainty over
the territory.’
(4)
Irgendwie finden sie es lustig, ein Stück über
Somehow consider they EXPL funny a play about
Revolution zu machen.
revolution to make
‘Somehow they see the fun in producing a play about
revolution.’
Of course, one could think of prototypical spatial senses
from which other senses are abstracted. It is thus not
improbable to argue that a HIERARCHICAL sense goes back to a
spatial relation that places one thing above or below another
(cf. [8]). But this does not make the preposition inherently
spatial, and a similar analysis cannot account for the
ABOUTNESS sense in (4) which then has to be assumed as a
second basic interpretation of über. Even if considering a
spatial sense as the basic interpretation of über, one will have
to decide which of the spatial senses should be the basic one.
Let us assume the projective one is the underlying sense. A
transit interpretation could be assumed as some kind of
abstraction: the traversing object passes the reference object’s
upside and thus can be located above the reference object. This
characterization becomes inconclusive, once one considers
examples like (5) in which the traversing object is not higher
than the traversed.
(5)
Interpretation
Sie wandern durch Tunnel über japanisches Gebiet.
‘They march through tunnels through Japanese
territory.’
What is more, it would be hard to explain why the transit
interpretation of über shows similarities to an interpretation of
durch (‘through’, ‘across’) (cf. section III.C for further
discussions) without assuming a projective vertical
interpretation as a basic interpretation of durch, too.
That a classification of preposition lexemes on the basis of
their senses is not necessarily useful can also be illustrated by
looking at a part of the prepositions that show a variety of other
interpretations in addition to their ‘prototypical’ spatial
interpretations, see Table a.
Given the huge diversity of interpretations of the
prepositions in Table a, it remains unclear why the prepositions
should be called spatial in the first place. Almost all of these
spatial prepositions exhibit a temporal interpretation, too. Why
The research reported herein has been supported by a grant of the
Deutsche Forschungsgemeinschaft (DFG) under KI-759/5.
temporal
causal
Prototypical spatial prepositions
an auf bei durch gegen hinter in neben über unter vor
• •
• •
•
conditional
•
•
•
•
•
•
•
•
•
•
medial
instrumental
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
• •
•
•
•
•
•
•
•
•
•
•
participation
correlation
•
•
statement
state
hierarchy
assignment
• •
•
•
order
•
•
transgression
agent
point of
reference
substitute
•
•
•
•
•
•
•
•
•
•
comparative
•
•
•
restrictive
•
copulative
•
•
pros & cons
adversative
•
•
distributive
initiation
•
•
theme
target
•
•
exchange
mereology
•
•
•
•
consecutive
circumstance
•
•
concessive
modal/manner
•
•
•
•
•
Table a: Prepositions with prototypical spatial interpretations and other senses
not call them temporal or perhaps even instrumental
prepositions? In addition, even if spatial senses are taken to be
primary or prototypical, this does not account for prepositions
that do not show spatial senses at all, or only in restricted
contexts. Consider the prepositions mit (‘with’) and ohne
(‘without’), which can be seen as antonyms in their sense
PARTICIPATION. While the first preposition can indeed have a
spatial interpretation in restricted contexts, viz. if the localized
object is in line with to the reference object, the second
preposition does not show a spatial sense at all.
Thus instead of assuming that spatial senses are primary or
prototypical, we assume that subsets of relational senses can be
assigned to prepositional lexemes, and that the presence of one
sense is no prima facie argument for the existence of another
for the same lexeme. This is what we call a relational analysis
of preposition senses. This relational view immediately
accounts for translation mismatches between closely related
languages. While English employs over for a projective vertical
interpretation, German über is assigned a projective vertical
interpretation, in addition to a sense of topological twodimensional transit, and a nonspatial interpretation of
ABOUTNESS. While the last is mapped to about in English, the
second interpretation has to be realized through across and
through.
interpretation is as adequate as a CAUSAL one. 4 In addition,
multiple annotations are applicable to ambiguities that cannot
be resolved without further context. This holds particularly if a
broader context is not provided, as e.g. in headlines. 5 The
sentence in (7) lacking a verb is ambiguous as to whether the
noun or the action is modified, meaning on the one hand that
the leader has no point and returns (PRESENCE) or on the other
hand that the leader returns without having made a point
(CIRCUMSTANCE).
(6)
The remaining paper is structured as follows: After some
words on the annotation scheme in general, the paper will focus
on the spatial interpretations of so-called spatial prepositions
(cf. (8)). We will particularly consider the role of established
terms like topological and projective, and point out the
difficulties encountered.
II.
(7)
GENERAL ARCHITECTURE OF THE ANNOTATION
SCHEME
The annotation scheme has been implemented in the
environment provided by the MMAX2 annotation tool (cf. [9])
and allows for an annotation in XML-stand-off format. We will
just briefly mention the most important properties of the
annotation scheme. An initial version of the scheme is
presented in [10], a broader introduction is given in [7].
One of the most important features of the annotation
scheme is its hierarchical organization. If senses can be
subcategorized into discernible subsenses, the subsenses are
structured in a classification tree that combines properties of a
taxonomical structure with that of a decision tree (as will be
illustrated with Figure 2 in section III.B). According to this
view, SPATIAL and TEMPORAL are supersenses with a rich and
diverse substructure. Given the incorporation of methods from
taxonomies and decision trees, the annotator either has to
choose between alternatives prescribed in the tree or has to
decide about the membership to one of the available
taxonomical classes. The temporal annotation tree (cf. [7]) is an
illustration of the first method. It is based on an analysis of
preposition choice, e.g. for language learners, in Durell and
Brée ([11]). Durrell and Brée wanted to offer guided choices
from senses to lexemes. Our task is to decide from a given
lexeme which sense is most plausible. We have appropriately
modified their decision tree and added some senses that have
not been covered by Durrell and Brée. The feasibility of the
temporal annotation scheme is being shown in an interannotator study, cf. [10].
‘The leader without a point back from Sion’
Before we turn to the analysis of spatial senses, we would
like to address a general feature of the annotation scheme: It
permits multiple annotations. Multiple annotations are useful in
cases where discernible interpretations interact or are closely
tied to one another. 3 See e.g. (6), where a TEMPORAL
Prepositions can be seen as multifunctional in these cases, just like it is
assumed for discourse markers (cf.[13]).
SPATIAL INTERPRETATIONS OF PREPOSITIONS
III.
In this section we present the part of our annotation scheme
that has been designed to account for spatial interpretations.
From the subset of 22 German prepositions, we have identified
those in (8) as prepositions that show a spatial interpretation
(the subset of so-called prototypical prepositions was listed in
Table a). Other prepositions that may have prototypical spatial
interpretations but are beyond that range will be left aside.
(8)
an, auf, bei, durch, gegen, hinter, in, mit, nach, neben,
über, um, unter, vor
Before introducing the annotation scheme to the reader, we
take a look at how traditional approaches classify spatial
prepositions. Kaufmann’s proposal (cf. [14]) clearly belongs to
this tradition and is the starting point for our analysis.
A. Traditional Classification of Prepositions with Spatial Use
spatial prepositions
simple localizations
(P[-DIR])
localization related to paths
target and source
prepositions
path prepositions
P[+DIR]
topological
With regard to spatial senses, we initially implemented
Schröder’s classification ([12]). Under closer scrutiny,
however, it turned out to be inapplicable, and consequently has
been replaced by the classification reported in the next chapter.
3
Grossfeuer nach Blitzschlag in Tramelan
(NZZ_1993_08_17_a135_seg9_s1, Headline: yes)
‘Serious fire in Tramelan after/because of lightning
stroke’
Der Leader ohne Punkt aus Sitten zurück
(NZZ_1994_03_14_a123_seg1_s1, Headline: yes)
in, an,
auf, bei
projective
topological
projective
vor, hinter,
unter, über
in, an,
aus, auf
vor, hinter,
unter, über
PV
CPV & MV
topological shape-related
(=Path-Ps) (=Route-Ps)
zu,
von
durch,
über
um,
längs/entlang
zu: CPV
& MV
von: MV
MV
MV, PV
um: +CPV
Figure 1. Classification of spatial prepositions (cf. [14])
4
Temporal and causal interpretations often go hand in hand, as the cause has
to precede the effect temporally.
5
We would like to thank an anonymous reviewer for suggesting a formal
distinction between multifunctionality and ambiguity. While the scheme does
not distinguish between these cases, indicating the distinction through an
additional feature could amend the annotation.
Kaufmann offers a classification for ‘spatial prepositions’
that is based on the prepositions’ combinatorial potential with
certain verb classes. As the verb classes are differentiated by
their combinatorial potential with different types of PPs with a
spatial use, the classification is in fact mutually dependent.
Three groups can be distinguished among verbs expressing a
localization: position verbs (PV) e.g. stand or sit, causative
position verbs (CPV) like put or place, and motion verbs (MV)
e.g. walk and fly.
The distinction applied to the prepositions is now made
between prepositions that encode simple localizations in a
region and those that describe localizations related to a path
(cf. Figure 1). The former class is further subdivided with
respect to its neighborhood region into topological (in, an, auf,
bei) and projective (vor, hinter, unter, über) prepositions, as it
is common in traditional literature. Topological prepositions
express relations between objects, while projective prepositions
“convey information about the direction in which an object is
related to the other” ([15]). The entire simple-localization class
is characterized by the fact that these prepositions can combine
with position verbs (9), but not denote the target of a motion
verb, as can be seen in (10).
(9)
(13) Er stellt die Tasche zu dem Tisch.
‘He puts the bag to the table.’
(14) Er läuft zu dem Tisch.
‘He runs to the table.’
(15) *Er legt die Tasche von dem Tisch zu der Theke.
‘He puts the bag from the table to the bar.’
(16) Er läuft von dem Tisch zu der Theke.
‘He runs from the table to the bar.’
The class that remains unexplained yet is that of the path
prepositions, comprising path prepositions that are topological
(durch, über), and route prepositions that are shape-related
(um, längs/entlang). The topological path prepositions describe
passages and can only be combined with motions verbs (17).
(17) Er geht durch das Gebäude.
‘He goes through the building.’
The prepositions that are shape-related hold the
characteristic that they are combinable with position and
motion verbs and the preposition um is additionally combinable
with causative position verbs.
(18) Sie sitzen um den Tisch/längs des Tisches.
‘They sit around/along the table.’
Er sitzt in/vor dem Café.
‘He sits in/in front of the café.’
(19) Sie laufen um den Tisch/längs des Tisches.
‘They run around/along the table.’
(10) *Er läuft in/vor dem Café.
‘He runs in/in front of the café.’ (Café = target)
The second class in Kaufmann’s taxonomy contains
prepositions expressing a localization related to a path. This
category branches into target and source prepositions and path
prepositions. The first subclass of target and source
prepositions is built of the prepositions expressing directionality. This branch is the directional equivalent to the prepositions expressing simple localizations. While prepositions
expressing simple localizations govern the dative case, the
target and source prepositions govern accusative case.
Correspondingly, this class is further differentiated into
topological (in, an, aus, auf) and projective (vor, hinter, unter,
über) directional prepositions. The directional prepositions are
subsumed here, as they combine with causative position (11)
and motion verbs (12). In this, they differ from prepositions
expressing simple localizations, which govern dative case, as
has been illustrated in (9) and (10).6
(11) Er stellt die Tasche in/vor das Café.
‘He puts the bag into/in front of the café.’
(12) Er läuft in/vor das Café.
‘He runs into/to the front of the café.’
The two prepositions von and zu are not classified as
directional. Instead they make up an individual class within the
category of target and source prepositions. While this class is
justified by the inability of von and zu to alternate between
dative and accusative case, homogeneity of the class cannot be
claimed, as zu combines with causative position and motion
verbs, while von only allows motion verbs.
(20) Sie stellen die Stühle um den Tisch/längs des Tisches.
‘They put the chairs around/along the table.’
A classification like the one discussed in this section could
serve as a starting point for an annotation of spatial preposition
senses. However, the spatial prepositions classified here do not
exactly match the prepositions we are interested in. Also, not
every preposition with a spatial interpretation was classified (it
does not cover gegen, mit, and nach) and, moreover, not every
spatial interpretation of the prepositions classified was taken
into consideration (for instance the following spatial subsenses
are not covered: a projective sense of auf as well as a
topological sense of unter and über7). It might be possible to
extend the presented classification to reach a more fine grained
classification and to add missing senses. With the given
arrangement of classifying attributes, we might encounter a
proliferate set of features which effectively blocks the
feasibility of the scheme for manual annotation.
Our goal is to classify preposition senses with high
granularity but with the concomitant aim for the classification
to be compact enough to serve the actual intention of manually
annotating senses with the help of the classification.
B. Mapping Prepositions to Spatial Senses
Instead of following the traditional view of classifying
spatial prepositions, the present scheme classifies spatial senses
that are associated with the respective prepositions. Traditional
criteria were used to distinguish different senses. A
rearrangement enables us to add missing interpretations as well
as adding more precise distinctions without inflating the
scheme, which is depicted in Figure 2.
6
A morphological differentiation between directional and static preposition,
like in English (e.g. in/into) does not exist in German.
7
See section III.D for the illustration of these senses.
unter
LO is
RO is
in the proximal area
in the boundary area
LO is
an, bei
an, auf, gegen
is exterior to RO
Dim(LO) is ? Dim (RO)
LO
area
über
layer
an, auf
pro-form for LO
(Stelle, Platz, Ort)
hidden by RO
hides RO
is contained in RO
less than or equal
greater than
in
auf
traverses RO
spatial
topological
passage of local
reference points
projective
shape-related
directed
nach
auf, über, unter
reference axes
vertical
horizontal 1
horizontal 2
traverses area
vertically related
to RO
Dim(LO) is ? Dim (RO)
less than or equal
greater than
durch
über
hinter, vor
über
neben
um
gegen, nach
target-orientated
relation LO to RO is
in line with/against
gegen, mit
Figure 2. Spatial tree with relations between LO and RO
Many preposition senses can be classified as topological in
locating the localized object (LO) in a neighboring region of
the reference object (RO). They have to be distinguished from
projective preposition senses, which locate the LO with the
help of reference axes as well as from shape-related um, one
special interpretation of nach and some directed senses.
Topological preposition senses can be split up further into the
ones locating the LO in a region inside the RO, the ones
locating the LO exterior to the RO, and the ones used for a
traversal of the RO by the LO (the transit interpretations). As
can be seen in Figure 2, the criteria for identifying the pertinent
senses for localizing a LO within a RO are identical to the ones
for localizing a traversal, except that they map to different
prepositions, an issue to which we will return in more detail
below.8
A feature in the scheme worth explaining is the topological
pro-form for the located object. It can be traced back to [16],
who observed that a concept like position (in German realized
as Stelle or Platz) does not refer to a concrete object but
functions as some kind of pro-form for local entities. The
reference of this concept can only be traced in a concrete
situation and with a given context.
The scheme in Figure 2 does not draw an obvious
distinction with respect to directionality. The reason is that
directionality does not affect the basic criteria for the
identification of spatial senses with regard to regions and axes.
Two aspects have to be considered here: First, a differentiation
between local and directional senses does not apply to every
spatial interpretation. Second, as was already pointed out in
section A, a case government alternation is correlated with the
assignment of directionality. A preposition with a static
localization sense typically governs the dative case, while a
directional interpretation leads to accusative case government,
as is illustrated in (21) and (22).
(21) Die Katze liegt vor dem Bett.
The cat lies in front of the bed.DAT
‘The cat lies in front of the bed.’
(22) Die Katze legt sich vor das Bett.
The cat lies down in front of the bed.ACC
‘The cat lies down in front of the bed.’
With regard to the first aspect, we note that the preposition
bei (‘at’, ‘by’, ‘near’), for example, never takes a complement
in the accusative 9 while über in its transit interpretation
(‘across’) is never followed by a dative10. Consequently, the
feature [±DIR] will be added to a sense after the classification
in Figure 2 has been traversed, and is not listed as a separate
feature in Figure 2 or Figure 3 below. The assignment is
applied automatically if directionality is correlated with a case
government alternation; and is set to a fixed value if only one
value is possible for a given sense.
In some spatial contexts the preposition nach
(‘after’/‘behind’) gets a spatial interpretation different from the
target-orientated one (‘to’) in (23). There seems to be an
interrelation with a temporal interpretation as well as with the
projective hinter that might be worth investigating further. We
call this relation in (24) a ‘passage of local reference points’ cf.
[12].
(23) Die Familie fährt nach Hamburg
‘The family drives to Hamburg.’
(24) Nach der Kapelle sind die Hügel zum Greifen nahe.
‘After/Behind the chapel the hills come within reach.’
We distinguish preposition senses that may receive the
feature directional from preposition senses that are directed.
Directed preposition senses express an inclination or alignment.
9
In most cases the directional counterpart of bei is zu (‘to’).
One could assume that all usages of über are directional. However,
examples like the following show that this assumption cannot be maintained:
1) Sie geht stundenlang über die Wiesen um Blumen zu suchen.
‘She walks across/over the meadows for hours, searching for flowers.’
10
8
The prepositions in Figure 2 are not part of the decision tree, but are listed
for illustrative purposes only.
auf
in, unter
borderline
crowd, collective
institution
RO is
meeting
an, auf, bei, vor
auf, bei, in, vor
travelling
(home of a) person
auf
working area
medium
bei
an
auf, in
Figure 3. Selectional restrictions on RO
They split up into the target-orientated interpretations of nach
and gegen (‘to’) ((25), (26)) and the interpretation ‘in line with’
of mit (‘with’) (27) and its counterpart gegen (‘against’) (28).
(25) Das Pendel schlug nach der Seite aus.
‘The pendulum swung to the side.’
(26) Das Pendel schlug gegen eine Seite aus.
‘The pendulum swung to the side.’
(27) Ernst fotografiert mit dem Licht.
‘Ernst takes a picture with the light.’
(28) Ernst fotografiert gegen das Licht.
‘Ernst takes a picture against the light.’
Within spatial senses, we distinguish localizations that can
be characterized with the help of the relation between LO and
RO (Figure 2) from localizations which seem to include
selectional restrictions on the syntactic object of the preposition
(Figure 3). The latter are typically excluded from systematic
classifications of spatial prepositions.
They cannot sufficiently be described by the relation
between LO and RO and require additional restrictions. There
is still a two-place relation established by the preposition but
some additional selectional restrictions on the object of the
preposition cause a change of the interpretation. A
compositional interpretation is no longer possible.
Take, e.g., the PP bei seinen Eltern in (29). The PP does not
denote a localization of Herbert in the proximal region of his
parents, but implies that he is visiting his parents’ home,
regardless of the parents being there at the same time.
Similarly, the PP am Fließband does not denote a localization
of its external argument in the proximal region of the assembly
line but indicates that he is working on the assembly line. This
interpretation shift is only possible if the object of the
preposition meets the relevant restrictions.11
11
Without a context the example (29) is ambiguous between an interpretation
in which the LO is localized in proximal region of the RO, and the
interpretation adhered to above, and thus would lend itself to multiple
annotations, as discussed in section II.
(29) Herbert ist bei seinen Eltern.
Herbert is at his parents
‘Herbert is at his parents’ home.’
(30) Er steht den ganzen Tag am Fließband.
He stands the whole day at the assembly line
‘He works on the assembly line the whole day.’
In some cases, apparent spatial senses of prepositions could
be traced back to prototypical localizations, but have departed
to a large degree from these, particularly since the prepositions
typically select an internal argument, which is an abstract entity
and cannot serve as a spatial reference object. We do not
classify such senses as SPATIAL interpretations; instead, the
scheme provides other senses, for example ASSIGNMENT that
can be assigned to the preposition in those examples.
(31) Er gab zu, dass der letzte Monat, mit fünfzehn
He admitted that the last month with fifteen
Todesopfern auf israelischer Seite, eine schwere
casualties on Israeli side a difficult
Periode gewesen sei.
period has been
‘He admitted that the last month has been a difficult
period with fifteen casualties on Israeli side.’
C. Relevant Dimensions; Inside and Through
It may astonish that we assume a sense of auf (‘on’) is
classified together with a sense of in as localizing the LO
within the RO. However, auf can be used as an alternative to in
in cases where the requirements of in concerning the
dimensions of LO and RO are not fulfilled. In requires its
internal argument, the RO, to have at least the number of
dimensions of the LO. To put it simply, a three-dimensional
object does not fit into a two-dimensional one (cf. [17]). One
could argue that in those examples auf offers a localization in a
boundary area of the RO or simply means ‘higher than
(+contact)’. Such an analysis, however, could not account for
the similarity of auf and in in the examples (32) and (33) nor
for the possibility of example (34). In this example the verb
vergraben (‘buried’) impedes an interpretation in which the LO
is located higher than the RO but only allows for an
interpretation of the LO being buried somewhere within the
area of the construction site.
(32) Es befanden sich Schweizer Truppen in deutschem
Gebiet.
‘There were Swiss troops in German territory.’
(33) Es befanden sich Schweizer Truppen auf deutschem
Gebiet.
‘There were Swiss troops on German territory.’
(34) Es lag auf der Baustelle vergraben.
‘It was buried on the construction site.’
Assuming auf as an alternative to in depending on the
dimensions of RO and LO (or the dimensions of their conceptualizations) but with roughly equal semantics also highlights
the consistent behavior of the path prepositions über and durch
(‘through’). Wherever auf is used for a localization inside the
RO (35), über is used for a traversal (36). Whenever the
requirements of in for a localization inside the RO are fulfilled
(37), for a traversal its analog durch is the right choice (38).
(35) Er liegt auf der Wiese.
‘He lies on the lawn.’
(36) Er geht über die Wiese.
‘He walks across the lawn.’
(37) Er liegt im Wald.
‘He lies in the woods.’
In addition to the aforementioned interpretations the
preposition über shows an interpretation referring to a traversal
of an area vertically related to the RO.13
(44) Er springt über die Mauer.
‘He jumps over the wall.’
We do not assume this interpretation to be projective
because in combination with a measurement it is not the height
that is measured but the distance of the goal from the RO or
even the space crossed by the jump.14
(38) Er geht durch den Wald.
‘He walks through the woods.’
D. Topological and Projective Interpretations of Prepositions
Reference [18] uses the combination with spatial
measurements like zwei Meter (‘two meters’) to distinguish
topological from projective prepositions. While we follow [18]
in employing spatial measurements as a criterion, we would
like to point out that it can only be a criterion to distinguish
topological from projective senses. Projective senses provide a
dimensional vector as a third argument in addition to LO and
RO, which can be identified with spatial measurements. For the
same reason, they do not find an anchor in topological senses,
which do not provide such an argument, as can be witnessed by
comparing the projective interpretations of vor and hinter in
(39) with the topological interpretations of bei and an in (40).12
(39) Zwei Meter vor/hinter dem Tisch
‘Two meters in front of/behind the table’
(40) *Zwei Meter bei/an dem Tisch
‘Two meters at/near the table’
While (39) and (40) might initially suggest that the
distinction hinges on the lexeme, i.e. that vor and hinter are
projective prepositions, but bei and an are topological, further
scrutiny shows that the distinction can indeed be applied to the
very same preposition; leading to complex sets of constraints
on localizations, as is witnessed for über in (41) and (42).
(41) Das Bild hängt über dem Loch.
The picture hangs above the hole
‘The picture hides the hole.’
(42) Das Bild hängt zwei Meter über dem Loch.
‘The picture hangs two meters above the hole.’
The interpretation of über changes depending on the
presence of a measurement. In the first sentences the hole is
(partially) hidden by the picture while in the second one the
picture is located two meters above the hole. The second one is
a projective use of the preposition while the first interpretation
is not. An arrangement in layers is relevant here instead, and
über systematically interchanges with unter depending on one
layer hiding the other (41) or being hidden by it (43).
(43) Das Loch ist unter dem Bild.
‘The hole is hidden by the picture.’
12
It should be noted, though, that the kind of measurement has to be taken into
consideration. Topological prepositions are not illicit with all spatial
measurements but only with numeral classifiers.
2) Nahe bei/an dem Tisch
‘Near by the table’
(45) Er springt 2m über die Mauer.
‘He jumps across the wall and lands 2m behind it.’
Conclusively it is not sufficient to say that über is a
projective preposition but that über has one interpretation
which is projective (this might be the more prominent one, but
it cannot be denied that the other interpretations exist).
Just as well, it is not satisfying to assume that auf is a
topological preposition. There are topological interpretations of
auf, like in (46), where no relevant directional vector or axis
can be detected.
(46) Die Schrift auf dem Schild
‘The lettering on the sign’
But to increase the complexity, auf installs some
restrictions concerning contact or support between the LO and
RO (e.g. [19], [12]). These preclude the use of measurements
defining a distance as a test for projective interpretations. From
the impossibility of (48) one cannot conclude that auf is not
projective but could argue that it is auf’s demand for contact
that forbids measurements.
(47) Die Tasse steht auf dem Tisch.
‘The cup stands on the table.’
(48) *Die Tasse steht zwei Meter auf dem Tisch.
‘The cup stands two meters on the table.’
Intuitively, an interpretation of sentences like (47) involves
the vertical axis. Building the opposite with unter (‘under’), auf
behaves like über with its projective interpretation and we
suggest assuming a projective interpretation of auf.
IV.
SUMMARY
We presented a new classification for spatial interpretations
of German prepositions that is part of a more comprehensive
annotation scheme for (an excerpt of) German prepositions.
In the classification we consider prototypical spatial senses
as well as some functional derivations of these senses that can
be defined by means of characteristics of the internal object of
the prepositions.
13
In these cases the RO has to have an extension on the vertical axis.
Verbs of motion offer the possibility to measure the length of the path,
irrespective the direction of the movement. But with projective prepositions
there should be an interpretation present that measures the actual distance on
the relevant axis.
3) Er geht drei Meter hinter den Baum.
≈He walks three meters (altogether) and stops somewhere behind the tree.
≈He walks and stops three meters behind the tree.
14
Instead of assigning defining attributes to a lemma of a
preposition we recommend a careful distinction between the
different interpretations of that preposition. As we have
evidenced, an interfusing of different interpretations results in
an incomplete picture of spatial prepositions.
[9]
[10]
REFERENCES
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
P. Saint-Dizier, “PrepNet: a Framework for Describing Prepositions:
Preliminary Investigation Results,” International Workshop on
Computational Semantics, Tilburg, The Netherlands, 2005.
K. Litkowski and O. Hargraves, “The Preposition Project,” Proc. ACLSIGSEM Workshop on “The Linguistic Dimensions of Prepositions and
Their Use in Computational Linguistic Formalism and Applications”,
University of Essex - Colchester, United Kingdom, 2005, pp. 171-179.
T. Kiss, K. Keßelmeier, A. Müller, C. Roch, T. Stadtfeld, and J. Strunk.
“A Logistic Regression Model of Determiner Omission in PPs,” Paper
for Proceedings of Coling 2010. Beijing, China, 2010.
H. Helbig, “Die semantische Struktur natürlicher Sprache,“ Berlin,
Heidelberg: Springer-Verlag, 2001.
M. Bierwisch, “On the Grammar of Local Prepositions,” In Syntax,
Semantik und Lexikon. Rudolf Ruzicka zum 65. Geburtstag, M.
Bierwisch, W. Motsch, and I. Zimmermann (eds.),Berlin: AkademieVerlag, 1988, pp. 1-65.
E. Lang, “The Meaning of German Projective Preposition: A Two-level
Approach,” in The Semantics of Prepositions. From Mental Processing
to Natural Language Processing, Zelinsky-Wibbelt,. Berlin/New York:
De Gruyter, 1993, pp. 249-291.
A. Müller, C. Roch, T. Stadtfeld, T. Kiss, “The Annotation of
Preposition Senses in German,” in Proceedings of Linguistic Evidence
2010, B. Stolterfoht and S. Featherston, Eds., Tübingen, Germany
Mouton de Gruyter, 2011, in press.
Deutsches Wörterbuch von Jakob und Wilhelm Grimm auf CD-Rom
und im Internet. © 1998-2004. Kompetenzzentrum für elektronische
Erschließungs- und Publikationsverfahren in den Geisteswissenschaften,
Trier.
(http://dwb.uni-trier.de/Projekte/WBB2009/DWB/wbgui_py?
lemid=GA00001).
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
C. Müller and M. Strube, “Multi-level annotation of linguistic data with
MMAX2,” in Corpus Technology and Language Pedagogy: New
Resources, New Tools, New Methods, S. Braun, K. Kohn, and J.
Mukherjee, Eds., Frankfurt am Main: Verlag Peter Lang GmbH, 2006,
pp. 197-214.
A. Müller, O. Hülscher, C. Roch, K. Keßelmeier, T. Stadtfeld, J. Strunk
and T. Kiss, “An Annotation Schema for Preposition Senses in
German,” in Proceedings of the ACL Linguistic Annotation Workshop
(LAW IV), Uppsala, Sweden, 2010.
M. Durell and D. Breé, “German temporal prepositions from an English
perspective,” in The Semantics of Prepositions. From Mental Processing
to Natural Language Processing, Cornelia Zelinsky-Wibbelt, Berlin/New
York: De Gruyter, 1993, pp. 295-325.
J. Schröder, “Lexikon deutscher Präpositionen“ VEB Verlag Enzyklopädie, 1986.
V. Petukhova and H. Bunt, “Towads a Multidimensional Semantics of
Dicourse Markers in Spoken Dialogue,” in Proceedings of the Eight
International Conference on Computational Semantics, pp.157-168.
I. Kaufmann, “Konzeptuelle Grundlagen semantischer Dekompositionsstrukturen. Die Kombinatorik lokaler Verben und prädikativer
Komplemente,” Tübingen: Max Niemeyer Verlag, 1995.
G. Retz-Schmitd, “Various Views on Spatial Prepositions,” AI
Magazine vol.9, No 2, 1988, pp.95-105.
J. Li, “Räumliche Relationen und Objektwissen am Beispiel an und bei,”
Studien zur deutschen Grammatik Band 49, Tübingen: Gunter Narr
Verlag, 1994.
I. Kaufmann, “Semantic and conceptual aspects of the preposition
durch,” in The Semantics of Prepositions. From Mental Processing to
Natural Language Processing. C. Zelinsky-Wibbelt, Edt. Berlin, New
York: Mouton de Gruyter, 1993, pp. 221-247.
D. Wunderlich, “Raum und Struktur des Lexikons,” in Perspektiven auf
Sprache. Interdisziplinäre Beiträge zum Gedenken an Hans Hörmann,
HG Bosshardt, Edt., Berlin:de Gruyter, 1986, pp. 212-231.
D. Wunderlich, “Lokale und Direktionale,” In: Semantik. Ein
internationales Handbuch zeitgenössicher Forschung, Vol. 6, A.von
Stechow, & D. Wunderlich, Eds. Berlin, New York: Walter de Gruyter,
1991, pp. 758-785.